Spontaneous, naturalistic behaviors comprise most of an animal’s lifetime. However, understanding how the brain encodes complex naturalistic behavior remains a fundamental challenge in neuroscience. Decades of research has shown that performance on overtrained, task-based, goal-oriented behaviors can be decoded from neural activity recorded in areas of the corticostriatal axis, namely primary motor cortex (M1) and dorsal striatum (DS). More recently, single-unit responses during the execution of trained behaviors were found to be stable over a timescale of weeks in M1. However, it is currently unknown whether untrained, naturalistic behaviors are stably represented by neurons across the corticostriatal axis. To address these questions, we simultaneously recorded single-unit activity in M1 and DS of freely-moving mice engaged in a range of untrained, spontaneous behaviors---including locomotion, grooming, and rearing---over the course of hours. We segmented naturalistic behaviors using B-SOiD, which identifies behavioral categories directly from patterns of animal pose and kinematics. While we found that a simple random forest model could predict which of the identified behaviors was being performed, the neural representation of behavior in M1 and DS drifted over a period of hours. Both single-unit activity patterns and population representations shifted significantly within-session despite kinematics remaining stable. We further assessed whether changes in the activity profiles of M1 neurons reflect changes in correlation structures across neurons or changes in low-dimensional dynamics. Using population dynamical systems models, we found that changes in single-neuron tuning comprise shifts in both low-dimensional dynamics and inter-neuron correlations. These results highlight key differences in how brains implement trained and naturalistic behavior. During skilled tasks, stable tuning and dynamics in M1 and DS coordinate actions. During unconstrained, naturalistic behaviors, a more fluid representation in both M1 and DS may govern how animals naturally interact with their environments.